1use csv::ReaderBuilder;
2use std::io::Read;
3use std::path::Path;
4
5use dataprof_core::{
6 ColumnProfile, DataProfilerError, DataSource, ExecutionMetadata, FileFormat, QualityDimension,
7 SemanticHints,
8};
9use dataprof_runtime::{
10 ProfileReport, ReportAssembler, StreamingColumnCollection, profile_builder,
11};
12
13mod memmap_reader;
14mod robust_csv;
15
16pub use memmap_reader::MemoryMappedCsvReader;
17pub use robust_csv::{
18 CsvDiagnostics, EncodingDiagnostic, RobustCsvParser, RobustParseResult, diagnose_non_utf8,
19};
20
21#[derive(Debug, Clone)]
26pub struct CsvParserConfig {
27 pub flexible: bool,
29 pub delimiter: Option<u8>,
31 pub quote_char: u8,
33 pub trim_whitespace: bool,
35 pub has_header: bool,
38 pub max_rows: Option<usize>,
40}
41
42impl Default for CsvParserConfig {
43 fn default() -> Self {
44 Self {
45 flexible: true,
46 delimiter: None,
47 quote_char: b'"',
48 trim_whitespace: false,
49 has_header: true,
50 max_rows: None,
51 }
52 }
53}
54
55impl CsvParserConfig {
56 pub fn strict() -> Self {
58 Self {
59 flexible: false,
60 ..Default::default()
61 }
62 }
63
64 pub fn with_delimiter(mut self, delimiter: u8) -> Self {
66 self.delimiter = Some(delimiter);
67 self
68 }
69
70 pub fn has_header(mut self, yes: bool) -> Self {
72 self.has_header = yes;
73 self
74 }
75
76 pub fn max_rows(mut self, max: Option<usize>) -> Self {
78 self.max_rows = max;
79 self
80 }
81}
82
83pub(crate) fn count_fields(line: &str, delimiter: char) -> usize {
85 let mut count = 1;
86 let mut in_quotes = false;
87 let mut chars = line.chars().peekable();
88 while let Some(ch) = chars.next() {
89 if ch == '"' {
90 if chars.peek() == Some(&'"') {
91 chars.next();
92 } else {
93 in_quotes = !in_quotes;
94 }
95 } else if ch == delimiter && !in_quotes {
96 count += 1;
97 }
98 }
99 count
100}
101
102fn split_sample_records(text: &str) -> (Vec<&str>, bool) {
110 let bytes = text.as_bytes();
111 let mut records = Vec::new();
112 let mut in_quotes = false;
113 let mut start = 0usize;
114 let mut i = 0usize;
115 while i < bytes.len() {
116 match bytes[i] {
117 b'"' => {
118 if in_quotes && bytes.get(i + 1) == Some(&b'"') {
119 i += 1; } else {
121 in_quotes = !in_quotes;
122 }
123 }
124 b'\n' if !in_quotes => {
125 let end = if i > start && bytes[i - 1] == b'\r' {
126 i - 1
127 } else {
128 i
129 };
130 records.push(&text[start..end]);
131 start = i + 1;
132 }
133 _ => {}
134 }
135 i += 1;
136 }
137 let last_unterminated = start < text.len();
138 if last_unterminated {
139 records.push(text[start..].trim_end_matches('\r'));
140 }
141 (records, last_unterminated)
142}
143
144pub fn detect_delimiter<R: Read>(reader: R) -> std::io::Result<u8> {
155 const SAMPLE_BYTES: u64 = 4096;
156 let mut preamble = Vec::new();
157 let mut take = reader.take(SAMPLE_BYTES);
158 take.read_to_end(&mut preamble)?;
159 let truncated = preamble.len() as u64 == SAMPLE_BYTES;
160
161 let preamble_str = String::from_utf8_lossy(&preamble);
162 let (mut records, last_unterminated) = split_sample_records(&preamble_str);
163 if truncated && last_unterminated && records.len() > 1 {
166 records.pop();
167 }
168
169 let delimiters = *b",;\t|";
170 let mut best_delimiter = b',';
171 let mut best_score = (false, 0usize, 0usize);
174
175 for &delimiter in &delimiters {
176 let mut counts = std::collections::HashMap::new();
177 for record in records.iter().take(5) {
178 let fields = count_fields(record, delimiter as char);
179 *counts.entry(fields).or_insert(0usize) += 1;
180 }
181
182 let (modal_fields, consistency) = counts
185 .into_iter()
186 .max_by_key(|&(fields, count)| (count, fields))
187 .unwrap_or((0, 0));
188
189 let score = (modal_fields > 1, consistency, modal_fields);
190 if score > best_score {
191 best_score = score;
192 best_delimiter = delimiter;
193 }
194 }
195
196 Ok(best_delimiter)
197}
198
199pub fn detect_delimiter_from_path(path: &Path) -> std::io::Result<u8> {
201 let file = std::fs::File::open(path)?;
202 detect_delimiter(std::io::BufReader::new(file))
203}
204
205pub fn analyze_csv_from_reader<R: Read>(
213 reader: R,
214 config: &CsvParserConfig,
215) -> Result<
216 (
217 Vec<ColumnProfile>,
218 StreamingColumnCollection,
219 usize,
220 Vec<String>,
221 ),
222 DataProfilerError,
223> {
224 analyze_csv_from_reader_with_hints(reader, config, &SemanticHints::default())
225}
226
227pub fn analyze_csv_from_reader_with_hints<R: Read>(
229 reader: R,
230 config: &CsvParserConfig,
231 semantic_hints: &SemanticHints,
232) -> Result<
233 (
234 Vec<ColumnProfile>,
235 StreamingColumnCollection,
236 usize,
237 Vec<String>,
238 ),
239 DataProfilerError,
240> {
241 let mut csv_builder = ReaderBuilder::new();
242 csv_builder.has_headers(config.has_header);
243 csv_builder.flexible(config.flexible);
244 csv_builder.quote(config.quote_char);
245 if config.trim_whitespace {
246 csv_builder.trim(csv::Trim::All);
247 }
248 let mut actual_delimiter = config.delimiter;
249
250 let boxed_reader: Box<dyn std::io::Read + '_> = if config.delimiter.is_none() {
251 let mut preamble = Vec::new();
252 let mut take = reader.take(4096);
253 take.read_to_end(&mut preamble)?;
254
255 let detected = detect_delimiter(std::io::Cursor::new(&preamble)).unwrap_or(b',');
256 actual_delimiter = Some(detected);
257
258 Box::new(std::io::Cursor::new(preamble).chain(take.into_inner()))
259 } else {
260 Box::new(reader)
261 };
262
263 if let Some(delimiter) = actual_delimiter {
264 csv_builder.delimiter(delimiter);
265 }
266
267 let mut csv_reader = csv_builder.from_reader(boxed_reader);
268
269 let header_names: Vec<String> = if config.has_header {
270 let headers = csv_reader.headers()?;
271 let names: Vec<String> = headers.iter().map(|header| header.to_string()).collect();
272 dataprof_core::validate_unique_column_names(&names, "CSV header")?;
275 names
276 } else {
277 let headers = csv_reader.headers()?;
278 (0..headers.len())
279 .map(|index| format!("column_{index}"))
280 .collect()
281 };
282
283 let mut column_stats = StreamingColumnCollection::new().with_semantic_hints(semantic_hints);
284 let mut rows_read = 0;
285
286 for result in csv_reader.records() {
287 if let Some(max_rows) = config.max_rows
288 && rows_read >= max_rows
289 {
290 break;
291 }
292
293 let record = result?;
294 let mut values: Vec<String> = record.iter().map(|value| value.to_string()).collect();
295
296 let header_len = header_names.len();
297 if values.len() < header_len {
298 values.resize(header_len, String::new());
299 } else if values.len() > header_len {
300 values.truncate(header_len);
301 }
302
303 column_stats.process_record(&header_names, values);
304 rows_read += 1;
305 }
306
307 let profiles = profile_builder::profiles_from_streaming_with_hints(
308 &column_stats,
309 false,
310 false,
311 None,
312 semantic_hints,
313 );
314
315 Ok((profiles, column_stats, rows_read, header_names))
316}
317
318pub fn analyze_csv_file(
323 file_path: &Path,
324 config: &CsvParserConfig,
325) -> Result<ProfileReport, DataProfilerError> {
326 analyze_csv_file_with_dimensions(file_path, config, None)
327}
328
329pub fn analyze_csv_file_with_dimensions(
331 file_path: &Path,
332 config: &CsvParserConfig,
333 quality_dimensions: Option<&[QualityDimension]>,
334) -> Result<ProfileReport, DataProfilerError> {
335 analyze_csv_file_with_dimensions_and_hints(
336 file_path,
337 config,
338 quality_dimensions,
339 &SemanticHints::default(),
340 )
341}
342
343pub fn analyze_csv_file_with_dimensions_and_hints(
344 file_path: &Path,
345 config: &CsvParserConfig,
346 quality_dimensions: Option<&[QualityDimension]>,
347 semantic_hints: &SemanticHints,
348) -> Result<ProfileReport, DataProfilerError> {
349 let metadata = std::fs::metadata(file_path).map_err(|error| map_io_error(file_path, error))?;
350 let start = std::time::Instant::now();
351
352 let file = std::fs::File::open(file_path).map_err(|error| map_io_error(file_path, error))?;
353 let buf_reader = std::io::BufReader::new(file);
354
355 let (column_profiles, column_stats, rows_read, header_names) =
356 analyze_csv_from_reader_with_hints(buf_reader, config, semantic_hints)?;
357
358 let file_source = DataSource::File {
359 path: file_path.display().to_string(),
360 format: FileFormat::Csv,
361 size_bytes: metadata.len(),
362 modified_at: None,
363 parquet_metadata: None,
364 };
365
366 if column_profiles.is_empty() && rows_read == 0 {
367 return Ok(ReportAssembler::new(
368 file_source,
369 ExecutionMetadata::new(0, header_names.len(), start.elapsed().as_millis())
370 .with_engine("csv"),
371 )
372 .skip_quality()
373 .build());
374 }
375
376 let sample_columns = profile_builder::quality_check_samples(&column_stats);
377 let scan_time_ms = start.elapsed().as_millis();
378 let num_columns = column_profiles.len();
379
380 let mut assembler = ReportAssembler::new(
381 file_source,
382 ExecutionMetadata::new(rows_read, num_columns, scan_time_ms).with_engine("csv"),
383 )
384 .columns(column_profiles)
385 .with_quality_data(sample_columns)
386 .with_row_duplicates(column_stats.row_duplicate_summary())
387 .with_exact_value_hint_bindings(column_stats.semantic_hint_bindings())
388 .with_semantic_hints(semantic_hints.clone());
389 if let Some(dimensions) = quality_dimensions {
390 assembler = assembler.with_requested_dimensions(dimensions.to_vec());
391 }
392 Ok(assembler.build())
393}
394
395fn map_io_error(file_path: &Path, error: std::io::Error) -> DataProfilerError {
396 if error.kind() == std::io::ErrorKind::NotFound {
397 DataProfilerError::FileNotFound {
398 path: file_path.display().to_string(),
399 }
400 } else {
401 DataProfilerError::from(error)
402 }
403}
404
405#[cfg(test)]
406mod tests {
407 use super::*;
408 use dataprof_core::DataType;
409 use std::io::{Cursor, Write};
410 use tempfile::NamedTempFile;
411
412 fn write_csv(content: &str) -> NamedTempFile {
413 let mut file = NamedTempFile::new().unwrap();
414 write!(file, "{}", content).unwrap();
415 file.flush().unwrap();
416 file
417 }
418
419 #[test]
420 fn test_analyze_csv_from_reader_basic() {
421 let data = b"name,age\nAlice,25\nBob,30\n";
422 let cursor = Cursor::new(data.as_ref());
423 let config = CsvParserConfig::default();
424
425 let (profiles, _stats, rows, _) = analyze_csv_from_reader(cursor, &config).unwrap();
426 assert_eq!(rows, 2);
427 assert_eq!(profiles.len(), 2);
428
429 let age = profiles
430 .iter()
431 .find(|profile| profile.name == "age")
432 .unwrap();
433 assert_eq!(age.total_count, 2);
434 assert_eq!(age.data_type, DataType::Integer);
435 }
436
437 #[test]
438 fn test_analyze_csv_from_reader_with_max_rows() {
439 let data = b"val\n1\n2\n3\n4\n5\n";
440 let cursor = Cursor::new(data.as_ref());
441 let config = CsvParserConfig::default().max_rows(Some(3));
442
443 let (_profiles, _stats, rows, _) = analyze_csv_from_reader(cursor, &config).unwrap();
444 assert_eq!(rows, 3);
445 }
446
447 #[test]
448 fn test_analyze_csv_from_reader_strict_rejects_ragged() {
449 let data = b"a,b\n1,2\n3,4,5\n";
450 let cursor = Cursor::new(data.as_ref());
451 let config = CsvParserConfig::strict();
452
453 assert!(analyze_csv_from_reader(cursor, &config).is_err());
454 }
455
456 #[test]
457 fn test_analyze_csv_from_reader_flexible_handles_ragged() {
458 let data = b"a,b\n1,2\n3,4,5\n6\n";
459 let cursor = Cursor::new(data.as_ref());
460 let config = CsvParserConfig::default();
461
462 let (profiles, _stats, rows, _) = analyze_csv_from_reader(cursor, &config).unwrap();
463 assert_eq!(rows, 3);
464 assert!(!profiles.is_empty());
465 }
466
467 #[test]
468 fn test_analyze_csv_from_reader_custom_delimiter() {
469 let data = b"a;b\n1;2\n3;4\n";
470 let cursor = Cursor::new(data.as_ref());
471 let config = CsvParserConfig::default().with_delimiter(b';');
472
473 let (profiles, _stats, rows, _) = analyze_csv_from_reader(cursor, &config).unwrap();
474 assert_eq!(rows, 2);
475 assert_eq!(profiles.len(), 2);
476 }
477
478 #[test]
479 fn test_analyze_csv_file_returns_quality_report() {
480 let csv = write_csv("x,y\n1,a\n2,b\n3,c\n");
481 let config = CsvParserConfig::default();
482 let report = analyze_csv_file(csv.path(), &config).unwrap();
483
484 assert_eq!(report.column_profiles.len(), 2);
485 assert_eq!(report.execution.rows_processed, 3);
486 assert!(report.quality_score().unwrap() >= 0.0);
487 assert!(report.quality_score().unwrap() <= 100.0);
488 }
489
490 #[test]
491 fn test_csv_auto_detects_delimiter_from_reader() {
492 let data = b"name;age\nAlice;25\nBob;30\n";
493 let cursor = Cursor::new(data.as_ref());
494 let config = CsvParserConfig::default();
495
496 let (profiles, _stats, rows, _) = analyze_csv_from_reader(cursor, &config).unwrap();
497 assert_eq!(rows, 2);
498 assert_eq!(profiles.len(), 2);
499
500 let age = profiles
501 .iter()
502 .find(|profile| profile.name == "age")
503 .unwrap();
504 assert_eq!(age.total_count, 2);
505 }
506
507 #[test]
508 fn test_analyze_csv_basic() {
509 let csv = write_csv("name,age\nAlice,25\nBob,30\n");
510 let config = CsvParserConfig::default();
511 let report = analyze_csv_file(csv.path(), &config).unwrap();
512 let profiles = &report.column_profiles;
513 assert_eq!(profiles.len(), 2);
514
515 let names: Vec<&str> = profiles
516 .iter()
517 .map(|profile| profile.name.as_str())
518 .collect();
519 assert!(names.contains(&"name"));
520 assert!(names.contains(&"age"));
521
522 let age = profiles
523 .iter()
524 .find(|profile| profile.name == "age")
525 .unwrap();
526 assert_eq!(age.total_count, 2);
527 assert_eq!(age.null_count, 0);
528 }
529
530 #[test]
531 fn test_analyze_csv_with_nulls() {
532 let csv = write_csv("name,age,email\nAlice,25,a@b.com\nBob,,\nCharlie,30,c@d.com\n");
533 let config = CsvParserConfig::default();
534 let report = analyze_csv_file(csv.path(), &config).unwrap();
535 let profiles = &report.column_profiles;
536
537 let age = profiles
538 .iter()
539 .find(|profile| profile.name == "age")
540 .unwrap();
541 assert_eq!(age.total_count, 3);
542 assert_eq!(age.null_count, 1);
543
544 let email = profiles
545 .iter()
546 .find(|profile| profile.name == "email")
547 .unwrap();
548 assert_eq!(email.null_count, 1);
549 }
550
551 #[test]
552 fn test_analyze_csv_file_returns_quality_report_legacy() {
553 let csv = write_csv("x,y\n1,a\n2,b\n3,c\n");
554 let config = CsvParserConfig::default();
555 let report = analyze_csv_file(csv.path(), &config).unwrap();
556
557 assert_eq!(report.column_profiles.len(), 2);
558 assert_eq!(report.execution.rows_processed, 3);
559 assert_eq!(report.execution.columns_detected, 2);
560 assert!(report.quality_score().unwrap() >= 0.0);
561 assert!(report.quality_score().unwrap() <= 100.0);
562 }
563
564 #[test]
565 fn test_analyze_csv_file_empty_file() {
566 let csv = write_csv("name,age\n");
567 let config = CsvParserConfig::default();
568 let report = analyze_csv_file(csv.path(), &config).unwrap();
569
570 assert_eq!(report.column_profiles.len(), 0);
571 assert_eq!(report.execution.rows_processed, 0);
572 }
573
574 #[test]
575 fn test_analyze_csv_with_max_rows_small_file() {
576 let csv = write_csv("val\n1\n2\n3\n4\n5\n");
577 let config = CsvParserConfig::default();
578 let report = analyze_csv_file(csv.path(), &config).unwrap();
579
580 assert_eq!(report.execution.rows_processed, 5);
581 assert_eq!(report.execution.columns_detected, 1);
582 assert!((report.execution.sampling_ratio.unwrap_or(1.0) - 1.0).abs() < 0.01);
583 }
584
585 #[test]
586 fn test_analyze_csv_numeric_types_detected() {
587 let csv = write_csv("int_col,float_col,str_col\n1,1.5,hello\n2,2.5,world\n3,3.5,foo\n");
588 let config = CsvParserConfig::default();
589 let report = analyze_csv_file(csv.path(), &config).unwrap();
590 let profiles = &report.column_profiles;
591
592 let int_col = profiles
593 .iter()
594 .find(|profile| profile.name == "int_col")
595 .unwrap();
596 assert!(matches!(int_col.data_type, DataType::Integer));
597
598 let float_col = profiles
599 .iter()
600 .find(|profile| profile.name == "float_col")
601 .unwrap();
602 assert!(matches!(float_col.data_type, DataType::Float));
603
604 let str_col = profiles
605 .iter()
606 .find(|profile| profile.name == "str_col")
607 .unwrap();
608 assert!(matches!(str_col.data_type, DataType::String));
609 }
610
611 #[test]
612 fn test_analyze_csv_strict_rejects_ragged_rows() {
613 let csv = write_csv("a,b\n1,2\n3,4,5\n");
614 let config = CsvParserConfig::strict();
615 assert!(analyze_csv_file(csv.path(), &config).is_err());
616 }
617
618 #[test]
619 fn test_analyze_csv_file_handles_ragged_rows() {
620 let csv = write_csv("a,b\n1,2\n3,4,5\n6\n");
621 let config = CsvParserConfig::default();
622 let report = analyze_csv_file(csv.path(), &config).unwrap();
623 assert!(!report.column_profiles.is_empty());
624 }
625
626 #[test]
627 fn test_analyze_csv_with_quality_metrics() {
628 let csv = write_csv(
629 "name,age\n\
630 Alice,25\nBob,\nCharlie,30\nDave,\nEve,28\n\
631 Frank,\nGrace,35\nHeidi,40\nIvan,\nJudy,22\n",
632 );
633 let config = CsvParserConfig::default();
634 let report = analyze_csv_file(csv.path(), &config).unwrap();
635
636 let age = report
637 .column_profiles
638 .iter()
639 .find(|profile| profile.name == "age")
640 .unwrap();
641 assert_eq!(age.null_count, 4);
642 assert_eq!(age.total_count, 10);
643 assert!(report.quality_score().unwrap() >= 0.0);
644 }
645
646 #[test]
647 fn test_detect_delimiter_comma() {
648 let data = b"name,age,city\nAlice,25,NYC\nBob,30,London\n";
649 assert_eq!(detect_delimiter(Cursor::new(data.as_ref())).unwrap(), b',');
650 }
651
652 #[test]
653 fn test_detect_delimiter_semicolon() {
654 let data = b"name;age;city\nAlice;25;NYC\nBob;30;London\n";
655 assert_eq!(detect_delimiter(Cursor::new(data.as_ref())).unwrap(), b';');
656 }
657
658 #[test]
659 fn test_detect_delimiter_pipe() {
660 let data = b"name|age|city\nAlice|25|NYC\nBob|30|London\n";
661 assert_eq!(detect_delimiter(Cursor::new(data.as_ref())).unwrap(), b'|');
662 }
663
664 #[test]
665 fn test_detect_delimiter_tab() {
666 let data = b"name\tage\tcity\nAlice\t25\tNYC\nBob\t30\tLondon\n";
667 assert_eq!(detect_delimiter(Cursor::new(data.as_ref())).unwrap(), b'\t');
668 }
669
670 #[test]
671 fn test_detect_delimiter_from_path_semicolon() {
672 let csv = write_csv("id;name;salary\n1;Alice;50000\n2;Bob;60000\n");
673 assert_eq!(detect_delimiter_from_path(csv.path()).unwrap(), b';');
674 }
675
676 #[test]
677 fn test_detect_delimiter_quoted_fields() {
678 let data = b"name;desc;val\n\"hello;world\";foo;1\nbar;baz;2\n";
679 assert_eq!(detect_delimiter(Cursor::new(data.as_ref())).unwrap(), b';');
680 }
681
682 #[test]
683 fn test_detect_delimiter_quoted_embedded_newline() {
684 let data = b"id,comment\n1,\"line one\nline two\"\n2,\"normal\"\n3,\"has \"\"quotes\"\" inside\"\n";
689 assert_eq!(detect_delimiter(Cursor::new(data.as_ref())).unwrap(), b',');
690 }
691
692 #[test]
693 fn test_detect_delimiter_absent_candidate_never_beats_real_one() {
694 let data = b"a,b,c\n1,2,3\nragged line without delimiters\n4,5,6\n";
697 assert_eq!(detect_delimiter(Cursor::new(data.as_ref())).unwrap(), b',');
698 }
699
700 #[test]
701 fn test_detect_delimiter_single_column_falls_back_to_comma() {
702 let data = b"value\n1\n2\n3\n";
703 assert_eq!(detect_delimiter(Cursor::new(data.as_ref())).unwrap(), b',');
704 }
705
706 #[test]
707 fn test_detect_delimiter_two_column_semicolon_with_commas_in_text() {
708 let data = b"id;note\n1;hello, world\n2;foo, bar, baz\n3;plain\n";
710 assert_eq!(detect_delimiter(Cursor::new(data.as_ref())).unwrap(), b';');
711 }
712
713 #[test]
714 fn test_split_sample_records_quote_aware() {
715 let (records, last_unterminated) = split_sample_records("a,b\r\n1,\"x\ny\"\n2,z\n");
716 assert_eq!(records, vec!["a,b", "1,\"x\ny\"", "2,z"]);
717 assert!(!last_unterminated);
718
719 let (records, last_unterminated) = split_sample_records("a,b\n1,\"cut mid-fie");
720 assert_eq!(records, vec!["a,b", "1,\"cut mid-fie"]);
721 assert!(last_unterminated);
722 }
723
724 #[test]
725 fn test_detect_delimiter_full_sample_ending_on_record_boundary() {
726 let mut data = Vec::from(&b"a;b;c\n"[..]);
729 while data.len() < 4090 {
730 data.extend_from_slice(b"1;2;3\n");
731 }
732 while data.len() < 4095 {
733 data.push(b'#');
734 }
735 data.push(b'\n');
736 assert_eq!(data.len(), 4096);
737 assert_eq!(
738 detect_delimiter(Cursor::new(data.as_slice())).unwrap(),
739 b';'
740 );
741 }
742}